VMware Virtual SAN best practices
This morning at VMworld 2014 there was a great session regarding the most common mistakes/pitfalls in VMware Virtual SAN design. One of the speakers was Cormac Hogan, VMware Storage Architect. He had some great tips.
- Minimal cluster size is 4 nodes. According to the requirements you can build a Virtual SAN using a 3 node cluster but you should take into account failure or maintenance of 1 node.
- Upgrading vSphere 5 to vSphere 5.5 Update 1 introduces a known bug. Install the fix mentioned in this VMware KB article.
- Use enterprise grade SSD. Selecting the SSD with the right performance profile can make a huge difference in VSAN performance easily. Both SSD and PCIe flash solutions are supported. All I/O Controllers, HDD and SSD must be on the Virtual SAN HCL. I hear people having great experience with Intel S3700 SSD.
- When adding new SSDs or HDDs ensure they are empty and not pre-formatted. Note: that when Virtual SAN is configured to “automatic mode” disks are added to existing disk groups or new disk groups are created automatically.
- Virtual SAN requires a SAS or SATA Host Bus Adapter (HBA), or RAID controller that is set up in non-RAID (pass through) mode.
- The Virtual SAN network must allow multicast traffic on the Virtual SAN network between the ESXi host participating in the Virtual SAN Cluster. When the Virtual SAN network does not allow multicast traffic, you will see that the network status is shown as being in a misconfigured state (Misconfiguration detected) after enabling Virtual SAN. You can ping all the Virtual SAN interfaces on all the hosts though.
- Use a disk controller with a queue depth > 256. When the queue depth is too low ,it can seriously impact the performance. A low queue depth will result in a lower performance. Why queue depth is so important? Check this article from Duncan Epping.
- Configure the SAS controller to disable non-local access. If you set your VSAN cluster to run in Automatic mode, you would expect it to claim all of the local disks. But some SAS controllers report their disks to ESXi as non-local because they allow their disks to be accessed by more than one host. If that’s not possible and Virtual SAN is not claiming your disks, go the Disk Management section and click on ‘Select all eligible disks’.
- Thin Provisioning is disabled by default. If you want to use thin provisioning, you wil have to enable it.
For more information one Virtual SAN and Virtual SAN Design.